Assessment of Fairness and Bias of an Image-based Surgical Site Infection Detection AI Model

The study evaluated the fairness of an AI model designed to detect surgical site infections (SSI) using wound images. The research included 13,702 images from 4,274 patients from 2019-2022, of which 9.8% had SSI. The model was tested for performance differences across age, gender, race, skin color, and geographic location. Results showed that the model performed fairly across all age, gender, race, and skin color groups with similar AUROC values (0.83-0.87). However, significant differences emerged in geographic location, with the lowest AUROC of 0.83 and highest of 0.87. Validation on new data from 2022-2023 confirmed that fairness was maintained across all groups except for geographic differences. The study emphasizes the importance of testing AI models for bias before their clinical deployment.